Feature Request: Support for Path objects in torchvision.datasets.ImageFolder

[songyuc]

🚀 The feature

It would be highly beneficial for PyTorch to support Path objects from the pathlib module when specifying dataset directories in functions like torchvision.datasets.ImageFolder. This support would make the code more Pythonic and in line with modern Python programming practices.

Motivation, pitch

Currently, when using torchvision.datasets.ImageFolder, the dataset directory must be specified as a string. However, the Path object is more versatile and is recommended for file system paths in Python 3. By supporting Path objects, users can take full advantage of the functionality provided by the pathlib module.

Example

from pathlib import Path
from torchvision import datasets, transforms

data_dir = Path("dataset", "train")
transform = transforms.Compose([
    transforms.Resize((256, 256)),
    transforms.ToTensor(),
    transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
])

# Currently, this requires converting Path to str
dataset = datasets.ImageFolder(str(data_dir), transform=transform)

# Proposed enhancement would allow this
dataset = datasets.ImageFolder(data_dir, transform=transform)

[NicolasHug]

Thanks for the feature request @songyuc . ImageFolder already supports Path objects, we probably just forgot to edit its type annotation. I've opened #8526 to fix that, it will be available in the next release soon, but you should already be able to use Path objects.